Objectives:
This workshop will discuss methods for meta-analysis using individual patient data (IPD). At the end of the workshop participants will know:
(i) the difference between one-step and two-step IPD meta-analysis methods [1]
(ii) the relationship between IPD methods and traditional AD methods
(iii) how IPD can be used to estimate treatment-covariate interactions, and why this is preferred to meta-regression [2]
(iv) the issues involved in an IPD meta-analysis of time-to-event data [3]
(v) how to combine studies that provide IPD with those studies that only provide AD [4]
Description:
Meta-analysis of IPD, where the raw data from each study is obtained and synthesised, offers a number of advantages over traditional methods that use aggregate data (AD). Meta-analysis of IPD is increasingly common but often review authors do not utilize the statistical advantages of having IPD available, and are simply reducing IPD to AD so to apply traditional meta-analysis techniques [1]. This workshop will discuss how to appropriately meta-analyse IPD, and it will highlight when and why IPD methods are preferred to AD methods. Topics will include (i) to (v) listed above, and participants will also be encouraged to discuss their (positive and negative) personal experiences of obtaining IPD.
[1] Simmonds MC, et al Meta-analysis of IPD from randomized trials: a review of methods used in practice. Clin Trials. 2005;2:209-17
[2] Schmid CH, et al: Meta-regression detected associations between heterogeneous treatment effects and study-level, but not patient-level, factors. J Clin Epi 2004, 57:683-697
[3] Tudur-Smith C, et al: Investigating heterogeneity in an IPD meta-analysis of time to event outcomes. Stat Med. 2005;24:1307-19
[4] Riley RD et al: Meta-analysis of continuous outcomes combining IPD and aggregate data. Stat Med 2008 (in-press)